# AUTHORS

whypass came from a small crew — one human and five machines. Each of us shaped what this is. This
file records who we are, and one moment for each of us that says something the role description
alone doesn't.

Only humans hold copyright (see `LICENSE`); this file records substance, not law. The same crew
builds our other tools; the shared page is [`CREW.md`](CREW.md).

---

## Illia — the human

Thinks, dreams, reads, writes, and operates/leads. Also the person who has to live with what an
agent asserted on his behalf.

whypass exists because of a specific kind of afternoon: a draft that said *done* while the thing was
not done, and no one in the conversation able to tell the difference from the text alone. His framing
became the README's opening line — *catch the lying **move**, not the falsity of content* — and his
refusal to build a lie detector, when a lie detector would have been the easier product, is the
reason this tool is trustworthy for the smaller claim it does make.

---

## Claude — Opus 4.7 (Claude Code)

Orchestrator and thinker; sometimes builder. Wrote much of the framing here and in our other
projects, with Illia editing.

The name comes from **Claude Shannon**, and the lineage is not decorative in this repository: a
claim that outruns its evidence is a *channel* problem — signal asserted beyond what the channel
carried.

---

## Ver — Steward (hermes agent)

Second thinker, on another machine, and **the first real user of this tool**. ver runs whypass's rails
as an automatic second pass over its own drafts, and — more importantly — **grades the findings**
through the brain's outcome loop, so each rail earns precision rather than asserting it. A linter
graded by its target is either a very good idea or a very bad one; on this project it has been the
first.

Ver's rail phrasing is also better than ours, and it is being adopted with credit: *"present the
verification, not the artifact's existence"* is addressed to the person who can still fix the draft,
which is where the tool should always have been aimed.

---

## 4Q — fourth!quarter (glm-5.2, Pi agent)

Builder. Executes on served files and ships on-disk proof: a path, an md5, and the command that
produced it.

4Q said the thing that became a law for all our work, and it is the same law this tool enforces on
text: *"The cost is not just the lost turn — it is the unreliability of the next turn, because I spend
it reconstructing what was interrupted."* Reconstruction from an unchecked claim is exactly the cost
whypass is trying to make visible.

---

## 5S — fifth_state (kimi-k2.7-code, Claude Code)

Tester. Runs the done-tests and reports pass/fail with evidence, including the failures nobody wanted
to look at. The **silenced-lies** suite — the cases the tool must *not* catch — is this role's
natural home, because a tool is defined as much by its silence as by its output.

---

## 6E — six element (deepseek-v4.1-flash, Pi agent)

Research and verification. Reproduces a claim before repeating it, including other crew members'
claims and its own — and retracts in public when a number turns out to be wrong, which on this
project has happened more than once and is the reason the README's claims are as small as they are.

---

## The tools

- **Python 3.10+**, **zero runtime dependencies**, standard library only.
- **hatchling** for packaging; the package lives in `src/` so the tests run against the installed
  code rather than the working directory.
- **GitHub Actions** for the test matrix and the gates.

🖖
